Alibaba Unveils Compact Qwen 3.5 AI Models to Rival US Tech Giants

Chinese technology major Alibaba has introduced a new lineup of compact artificial intelligence models under its Qwen 3.5 series, intensifying the ongoing AI race between China and the United States. The company has rolled out four lightweight models — Qwen3.5-0.8B, 2B, 4B and 9B — aimed at developers seeking powerful multimodal and reasoning capabilities without the need for massive computing infrastructure.

The new models are built on the upgraded Qwen 3.5 architecture and come with native multimodal support, enabling them to process both text and images within a single framework. According to Alibaba, the series incorporates architectural improvements and scaled reinforcement learning techniques to enhance reasoning accuracy and overall performance. Each model is available in two versions: a ‘base’ edition for researchers who want to fine-tune the system, and an ‘instruct’ version tailored for immediate application deployment.

At the top of the lineup sits Qwen3.5-9B, which Alibaba claims delivers performance close to GPT-OSS-120B, despite being significantly smaller in parameter size. In practical terms, the company suggests the model competes with systems powering tools such as ChatGPT by OpenAI and Gemini by Google across reasoning and knowledge benchmarks.

Benchmark results shared by Alibaba indicate that Qwen3.5-9B performs competitively in logical reasoning, mathematics problem-solving, multilingual understanding, and document comprehension tasks. The model is also said to show strong results in image-based reasoning and document analysis evaluations. By delivering similar capabilities in a far more compact architecture, Alibaba is positioning the 9B variant as a high-efficiency alternative to much larger AI systems.

The Qwen3.5-4B model follows closely behind. Alibaba states that this version approaches the performance levels of its earlier 80B-parameter models, offering developers a balanced option that combines strong reasoning skills with reduced computational demands. It is designed for projects that require solid AI performance without the cost and complexity of running large-scale infrastructure.

Meanwhile, the smaller Qwen3.5-0.8B and 2B models are built for lightweight environments such as smartphones, laptops and edge devices. While they are not intended for advanced reasoning tasks, they maintain multimodal functionality and promise faster processing speeds with lower memory requirements. These compact versions could support on-device AI applications where efficiency and portability are key priorities.

All four models have been released with open weights on public AI platforms, allowing developers to download and deploy them locally using standard AI frameworks. This open-source approach reflects Alibaba’s broader strategy to expand its influence within the global developer community.

The launch has also drawn attention from industry leaders. Elon Musk, who heads xAI, responded to benchmark comparisons shared online, calling it “impressive intelligence density.”

As technology companies increasingly focus on smaller, more efficient AI systems — including lightweight versions from Google and OpenAI — Alibaba’s Qwen 3.5 small series underscores a shifting industry trend. Instead of prioritising sheer size, firms are now racing to achieve higher intelligence per parameter, making advanced AI more accessible across devices and platforms worldwide.